9 SCIENTIFIC HIGHLIGHT OF THE MONTH

Exchange interactions, spin waves, and

transition temperatures in itinerant magnets
I. Turek,
1,2,5
J. Kudrnovsk y,
3
V. Drchal,
3
and P. Bruno
4
1
Institute of Physics of Materials, Academy of Sciences of the Czech Republic,

Zizkova 22, CZ-61662 Brno, Czech Republic

2
Department of Electronic Structures, Charles University in Prague,
Ke Karlovu 5, CZ-12116 Prague 2, Czech Republic
3
Institute of Physics, Academy of Sciences of the Czech Republic,
Na Slovance 2, CZ-18221 Prague 8, Czech Republic
4
Max-Planck-Institut f ur Mikrostrukturphysik, Weinberg 2, D-06120 Halle, Germany
5
Center for Computational Materials Science, Technical University of Vienna,
Getreidemarkt 9/158, A-1060 Vienna, Austria
Abstract
The contribution reviews an ab initio two-step procedure to determine exchange inter-
actions, spin-wave spectra, and thermodynamic properties of itinerant magnets. In the
rst step, the selfconsistent electronic structure of a system is calculated for a collinear spin
structure at zero temperature. In the second step, parameters of an eective classical Heisen-
berg Hamiltonian are determined using the magnetic force theorem and the one-electron
Green functions. The Heisenberg Hamiltonian and methods of statistical physics are em-
ployed in subsequent evaluation of magnon dispersion laws, spin-wave stiness constants,
and Curie/Neel temperatures. Applicability of the developed scheme is illustrated by se-
lected properties of various systems like transition and rare-earth metals, disordered alloys
including diluted magnetic semiconductors, ultrathin lms, and surfaces.
1 Introduction
The quantitative description of ground-state and nite-temperature properties of metallic sys-
tems represents a long-term challenge for solid state theory. Practical implementation of density
functional theory (DFT) [1, 2, 3] led to excellent parameter-free description of ground-state
112
properties of metallic magnets, including traditional bulk metals and ordered alloys as well as
systems without the perfect three-dimensional periodicity, like, e.g., disordered alloys, surfaces
and thin lms. On the other hand, an accurate quantitative treatment of excited states and
nite-temperature properties of these systems remains an unsolved problem for ab initio theory
[4, 5, 6, 7] despite the formal extension of the DFT to time-dependent phenomena [8] and -
nite temperatures [9]. The usual local spin-density approximation (LSDA) [3] fails to capture
important features of excited states, in particular the magnetic excitations responsible for the
decrease of the magnetization with temperature and for the magnetic phase transition.
In developing a practical parameter-free scheme for the nite-temperature magnetism, one has
to rely on additional assumptions and approximations the validity of which has to be chosen
on the basis of physical arguments. The purpose of this contribution is to review theoretical
backgrounds, numerical aspects, and selected results of an approach formulated nearly two
decades ago [10, 11] (see Ref. [12] for a recent review), and applied by the present authors
to a number of qualitatively dierent systems [13, 14, 15, 16, 17, 18, 19, 20], including also
yet unpublished results. The review is organized as follows: Section 2 lists the underlying
physical concepts and approximations of the scheme and Section 3 deals with computational
details and specic problems related to its numerical implementation. Examples of applications
are given in Section 4: bulk transition metals (Section 4.1), rare-earth metals (Section 4.2),
disordered alloys (Section 4.3), diluted magnetic semiconductors (Section 4.4), two-dimensional
ferromagnets (Section 4.5), and surfaces of bulk ferromagnets (Section 4.6). Comparisons to
other authors using the same (or similar) approach are made throughout Section 4, while a
critical discussion of the scheme and a brief comparison to alternative approaches are left to the
last section (Section 5).
2 Formalism
It is well known that magnetic excitations in itinerant ferromagnets are basically of two dierent
types, namely, the Stoner excitations, in which an electron is excited from an occupied state of
the majority-spin band to an empty state of the minority-spin band and creates an electron-hole
pair of triplet spin, and the spin-waves, or magnons, which correspond to collective transverse
uctuations of the magnetization direction. Near the bottom of the excitation spectrum, the
density of states of magnons is considerably larger than that of corresponding Stoner excitations
(associated with longitudinal uctuations of the magnetization), so that the thermodynamics
in the low-temperature regime is completely dominated by magnons and Stoner excitations
can be neglected. Therefore it seems reasonable to extend this approximation up to the Curie
temperature and to derive an ab initio technique of nite-temperature magnetism by neglecting
systematically the Stoner excitations.
With thermodynamic properties in mind, we are primarily interested in the long-wavelength
magnons with the lowest energy. We adopt the adiabatic approximation [21] in which the
precession of the magnetization due to a spin-wave is neglected when calculating the associated
change of electronic energy. The condition of validity of this approximation is that the precession
time of the magnetization should be large as compared to characteristic times of electronic
113
motion, i.e., the hopping time of an electron from a given site to a neighboring one and the
precession time of the spin of an electron subject to the exchange eld. In other words, the spin-
wave energies should be small as compared to the band width and to the exchange splitting. This
approximation becomes exact in the limit of long wavelength magnons, so that the spin-wave
stiness constants calculated in this way are in principle exact.
This procedure corresponds to evaluation of changes of the total energy of a ferromagnet due to
innitesimal changes of the directions of its local magnetic moments associated with individual
lattice sites R. The directions of the moments are specied by unit vectors e
R
. An exact
calculation of the total energy ce
R
of a prescribed spin conguration leads to the constrained
density functional theory [22], which allows to obtain the ground state energy for a system subject
to certain constraints. The latter are naturally incorporated into the DFT in terms of Lagrange
multipliers. In the present case, the constraint consists in imposing a given conguration of
spin-polarization directions, namely, along e
R
within the atomic (Wigner-Seitz) cell R. The
Lagrange multipliers can be interpreted as magnetic elds B

R
constant inside the cells with
directions perpendicular to the unit vectors e
R
. Note that intracell non-collinearity of the
spin-polarization is neglected since we are primarily interested in low-energy excitations due to
intercell non-collinearity. In the so-called frozen-magnon approach, one chooses the constrained
spin-polarization conguration to be the one of a spin-wave with the wave vector q and computes
the spin-wave energy E(q) directly by employing the generalized Bloch theorem for a spin-spiral
conguration [23].
In a real-space approach, adopted here, one calculates directly the energy change associated with
a constrained rotation of the spin-polarization axes in two cells e
R
and e
R
. This represents
a highly non-trivial task which requires selfconsistent electronic structure calculations for non-
collinear spin-polarized systems without translational periodicity. Restriction to innitesimal
changes of the moment directions, u
R
= e
R
e
0
, perpendicular to the direction of the ground-
state magnetization e
0
, leads to an expansion of ce
R
to second order in u
R
of the form
[11, 24]
cu
R
=

RR

A
RR
u
R
u
R
. (1)
This expression can be extended to nite changes of the moment directions using an eective
Heisenberg Hamiltonian (EHH)
H
e
e
R
=

RR

J
RR
e
R
e
R
. (2)
The constants J
RR
in Eq. (2), the pair exchange interactions, are parameters of the EHH which
satisfy J
RR
= J
R

R
and J
RR
= 0. They are related to the coupling constants A
RR
of Eq. (1)
by
A
RR
= J
RR
+
RR

J
R

R
_
(3)
so that an important sum rule

R
A
RR
=

R

A
RR
= 0 (4)
is satised which guarantees that the total energy remains invariant upon a uniform rotation of
the magnetization.
114
The practical calculations of the exchange interactions J
RR
in ferromagnets are greatly sim-
plied by using the magnetic force theorem [10, 11] (a similar approach was also suggested for
magnetic impurities in a non-magnetic host [25]). The innitesimal changes of the total energy,
Eq. (1), can be expressed using changes in one-particle eigenvalues due to non-selfconsistent
changes of the eective one-electron potential accompanying the innitesimal rotations of spin
quantization axes, i.e., without any additional selfconsistent calculations besides that for the
collinear ground state. The resulting pair exchange interactions are given by [11, 24]
J
RR
=
1

Im
_
E
F

dE
_

R
dr
_

dr

B
xc
(r) G

(r, r

; E + i0) B
xc
(r

) G

(r

, r; E + i0) , (5)
where E
F
denotes the Fermi level,
R
denotes the R-th atomic cell, B
xc
(r) is the exchange-
correlation magnetic eld, 2B
xc
(r) = V

(r) V

(r), where V

(r) ( =, ) is the selfconsistent

LSDA potential, and G

(r, r

; E + i0) is the one-electron retarded Green function for the same

potential. It should be noted that the parameters J
RR
determined by Eq. (5) do not con-
tain contributions due to constraining magnetic elds necessary to keep a frozen non-collinear
spin structure a stationary state of the Kohn-Sham equation. It can be expected that these
contributions can be neglected in systems with large local magnetic moments. Validity of this
approximation have been put on a more quantitative level in recent studies [24, 26].
Once the exchange parameters J
RR
are obtained, the adiabatic spin-dynamics [27, 28, 29, 30, 31]
can be easily determined from the EHH, Eq. (2). One obtains the result known from spin-wave
theories of localized ferromagnets: for ferromagnetic crystals with one atom in the primitive cell,
the energy E(q) of a zero-temperature magnon is related to the lattice Fourier transform J(q)
of the exchange interactions J
RR
by
E(q) =
4
M
[J(0) J(q)] , J(q) =

R
J
0R
exp(iq R) , (6)
where q denotes a vector in the Brillouin zone (BZ) of the lattice, M denotes the local moment
magnitude in units of
B
(
B
is the Bohr magneton). For cubic systems and for small q-vectors,
E(q) D[q[
2
with the spin-wave stiness constant equal to
D =
2
3M

R
[R[
2
J
0R
. (7)
Finally, to obtain thermodynamic quantities such as the Curie temperature, methods of statis-
tical mechanics have to be applied to the EHH, Eq. (2). The simplest estimate of the Curie
temperature is based on a mean-eld approximation (MFA) which leads to
k
B
T
MFA
C
=
2
3
J
0
, J
0
=

R
J
0R
= J(0) , (8)
where k
B
is the Boltzmann constant. The quantity J
0
(on-site exchange parameter) reects
the molecular eld experienced by a single moment in the ferromagnet. The limitations of the
MFA are well known: it is correct only in the limit of high temperatures (above T
C
) and it fails
to describe the low-temperature collective excitations (spin-waves). An improved description
of nite-temperature properties is provided by the Green function method within the random
phase approximation (RPA) [32, 33]. The RPA is valid not only for high temperatures, but
115
also at low temperatures, and it describes correctly the spin-waves. In the intermediate regime
(around T
C
), it represents a rather good approximation which may be viewed as an interpolation
between the high and low temperature regimes. The RPA formula for the Curie temperature is
given by
_
k
B
T
RPA
C
_
1
=
3
2
1
N

q
_
J
0
J(q)
_
1
, (9)
where N denotes the number of q-vectors used in the BZ-average. It can be shown that T
RPA
C
is
always smaller than T
MFA
C
. It should be noted, however, that both the MFA and the RPA fail
to describe correctly the critical behavior and yield in particular incorrect critical exponents.
Finally, the Curie temperature can also be estimated purely numerically by employing the
method of Monte Carlo simulations applied to the EHH. This approach is in principle exact but
its application to real itinerant systems requires inclusion of a sucient number of neighboring
shells due to long-ranged interactions J
RR
(see Section 3.2).
3 Numerical implementation
3.1 Selfconsistent electronic structure
Ecient evaluations of the pair exchange interactions, Eq. (5), require a rst-principle tech-
nique which provides the one-electron Green function in the real space. The results reported
here are based on selfconsistent LSDA calculations using the all-electron non-relativistic (scalar-
relativistic) tight-binding linear mun-tin orbital (TB-LMTO) method and the atomic-sphere
approximation (ASA) [34, 35, 36], with the exchange-correlation potential parametrized accord-
ing to Ref. [37]. The energy integrals over the occupied part of the valence band were expressed
as integrals over an energy variable along a closed path C starting and ending at the Fermi en-
ergy (with the occupied part of the valence band lying inside C). The integrals were numerically
evaluated using the Gaussian quadrature method [35, 36]. Other Green function techniques, es-
pecially the the Korringa-Kohn-Rostoker (KKR) method [38, 39], are equally suitable in the
present context.
Within the ASA, the Green function for a closely packed solid can be written in the form [35, 40]
(the spin index is omitted for brevity in Section 3.1)
G(r +R, r

+R

; z) =
RR

RL
(r
<
, z)
RL
(r
>
, z)
+

LL

RL
(r, z) G
RL,R

L
(z)
R

L
(r

, z) . (10)
In Eq. (10), the variables r, r

refer to positions of points inside the individual atomic spheres,

z denotes a complex energy variable, the symbol r
<
(r
>
) denotes that of the vectors r, r

with
the smaller (larger) modulus, and L, L

are the angular momentum indices, L = (, m). The

functions
RL
(r, z) and
RL
(r, z) denote, respectively, properly normalized regular and irregular
solutions of the Schr odinger equation for the spherically symmetric potential inside the R-
th atomic sphere. All multiple-scattering eects are contained in the Green function matrix
G
RL,R

L
(z) which is given in terms of the potential functions P
R
(z) and the structure constants
116
S
RL,R

L
of the LMTO method by
G
RL,R

L
(z) =
R
(z)
RL,R

L
+
R
(z) g
RL,R

L
(z)
R

(z) , (11)
where the quantities on the r.h.s. are dened as

R
(z) =
_

P
R
(z) ,
R
(z) =
1
2

P
R
(z)

P
R
(z)
,
g
RL,R

L
(z) =
_
[P(z) S]
1
_
RL,R

. (12)
In the last equation, the symbol P(z) denotes a diagonal matrix of potential functions dened as
P
RL,R

L
(z) = P
R
(z)
RL,R

L
and an overdot means energy derivative. The matrix g
RL,R

L
(z)
will be referred to as the auxiliary (or KKR-ASA) Green function. The quantities P
R
(z),

R
(z),
R
(z), S
RL,R

L
, and g
RL,R

L
(z) can be expressed in any particular LMTO represen-
tation (canonical, screened); the resulting Green function matrix, Eq. (11), the Green function,
Eq. (10), and all derived physical quantities are invariant with respect to this choice. However,
the most screened (tight-binding) representation is the best suited for most calculations and
it has been employed in the present implementation. The energy dependence of the potential
functions P
R
(z) is parametrized in terms of three standard potential parameters, i.e., with the
second-order accuracy [34, 35].
3.2 Parameters of the classical Heisenberg Hamiltonian
Substitution of the Green function G

(r, r

; z) in the ASA (Section 3.1) into Eq. (5) yields an

expression suitable for computations [11, 12, 14], namely,
J
RR
=
1
8i
_
C
tr
L
_

R
(z) g

RR

(z)
R
(z) g

R
(z)
_
dz ,

R
(z) = P

R
(z) P

R
(z) , (13)
where tr
L
denotes the trace over the angular momentum index L and energy integration is
performed along the contour C described in Section 3.1. The quantities g

RR
(z) ( =, ) denote
site-o-diagonal blocks of the auxiliary Green-function matrices with elements g

RL,R

L
(z) while

R
(z) are diagonal matrices related to the potential functions P

R
(z). The diagonal elements
of
R
(z) play a role of energy- and -dependent exchange splittings on individual atoms while
the expression (13) for the exchange interactions J
RR
has a form of a bare static transversal
susceptibility.
Well converged calculations of the exchange interactions J
RR
for bulk metals with perfect
translational symmetry for distances d = [R R

[ up to ten lattice constants a require high

accuracy of the full BZ-averages (inverse lattice Fourier transforms) dening the site-o-diagonal
blocks g

RR
(z) [35, 36]. In particular, we have used typically a few millions of k-points in the
full BZ for the energy point on the contour C closest to the Fermi energy, and the number of k-
points then progressively decreased for more distant energy points [14, 17]. A typical evaluation
of exchange interactions requires about two hours on P4-based personal computers.
The calculated Heisenberg exchange parameters for bcc Fe (with experimental value of its lattice
constant) are shown in Fig. 1. One can see dominating ferromagnetic interactions for the rst
117
and second nearest-neighbor shells followed by weaker interactions of both signs and decreasing
magnitudes for bigger distances d = [RR

[ (Fig. 1, left panel). The same qualitative features

were found for other 3d ferromagnets: fcc Co, fcc Ni [14] and hcp Co [18].
-0.5
0.0
0.5
1.0
1.5
2.0
1 2 3 4 5 6 7 8 9 10
J


(
m
R
y
)
d/a
Fe bcc
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
1 2 3 4 5 6 7 8 9 10
(
d
/
a
)
3

J


(
m
R
y
)
d/a
Fe bcc
Figure 1: Exchange interactions J
RR
for bcc Fe as a function of the distance [R R

[ = d
without (left panel) and with (right panel) a prefactor d
3
.
An analysis of the exchange interactions J
RR
, Eq. (13), in the limit of large distances d =
[R R

[ has been given in Ref. [14] for a a single-band model using the stationary-phase ap-
proximation [41]. For a weak ferromagnet, one reveals a characteristic Ruderman-Kittel-Kasuya-
Yoshida (RKKY) asymptotic behavior
J
RR

sin
__
k

F
+k

F
_
(RR

) +
_
[RR

[
3
, (14)
where k

F
is a Fermi wave vector in a direction such that the associated group velocity is parallel
to R R

, and denotes a phase factor. The exchange interaction according to Eq. (14) has
an oscillatory character with an envelope decaying as [R R

[
3
. On the other hand, for a
strong ferromagnet with a fully occupied majority band the corresponding Fermi wave vector is
imaginary, namely, k

F
= i K

F
, and one obtains an exponentially damped RKKY behavior
J
RR

sin
_
k

F
(RR

) +
_
exp
_
K

F
(RR

)
_
[RR

[
3
. (15)
The qualitative features of these RKKY-type oscillations will not be changed in realistic ferro-
magnets. This is illustrated for bcc Fe (weak ferromagnet) in Fig. 1 (right panel) which proves
undamped oscillations of the quantity [R R

[
3
J
RR
. It should be noted that due to the sp-d
hybridization no itinerant ferromagnet is a truly strong ferromagnet the only exceptions are
half-metallic ferromagnets.
3.3 Magnetic properties from the Heisenberg Hamiltonian
The RKKY-like asymptotic behavior, Eq. (14), leads to numerical diculties in calculations of
the magnon spectra and the spin-wave stiness constants. The lattice Fourier transform of the
118
exchange interactions, Eq. (6), is not an absolutely convergent sum and its convergence with
respect to the number of shells included has to be carefully checked (see Section 4.1). Note,
however, that the lattice sum over [J
0R
[
2
does converge so that J(q) is dened unambiguously
in the L
2
sense.
0 2 4 6 8
d
max
/a
0
200
400
600
800
1000
D

[
m
e
V
.
A
2
]
= 0.0
= 0.2
= 0.4
= 0.6
= 0.8
Ni
Figure 2: Spin-wave stiness of fcc Ni as a function of d
max
(in units of lattice constant) for
various values of the damping factor .
The lattice sum for the spin-wave stiness constant, Eq. (7), is not convergent at all, and the
values of D as functions of a cut-o distance d
max
exhibit undamped oscillations for all three
cubic 3d ferromagnets [14]. To resolve this diculty we suggested to regularize the original
expression, Eq. (7), by replacing it by the formally equivalent expression which is, however,
numerically convergent
D() =
2
3M
lim
dmax

|R|<dmax
[R[
2
J
0R
exp([R[/a) ,
D = lim
0
D() , (16)
where a is the lattice constant. The quantity plays a role of a damping parameter which makes
the lattice sum absolutely convergent as it is seen from Fig. 2 for the case of fcc Ni.
It can be shown that the quantity D() is an analytical function of the variable for any value
> 0 and it can be extrapolated to the value = 0. We therefore perform calculations for a set
of values (
min
,
max
) for which D() is a smooth function with a well dened limit for large
d
max
. The limit 0 is then determined at the end of calculations by a quadratic least-square
extrapolation method. The procedure is illustrated in Fig. 3 for the cubic Fe, Co and Ni. Note
that these convergence problems are less serious in half-metallic magnets due to the exponential
damping described by Eq. (15).
Direct calculations of the Curie temperatures in the MFA according to Eq. (8) face convergence
problems similar to the magnon spectra. Alternatively, one can evaluate the on-site exchange
parameter J
0
using a sum rule valid also for systems without translational periodicity [11]:
J
0
R
=

R

J
RR
=
1
8i
_
C
tr
L
_

R
(z)
_
g

RR
(z) g

RR
(z)
_
119
0 0.2 0.4 0.6 0.8 1

0
200
400
600
800
D

[
m
e
V
.
A
2
]
Fe
Ni
Co
Figure 3: Spin-wave stiness coecients D() for bcc Fe, fcc Co, and fcc Ni as a function of the
parameter (open symbols) and extrapolated values for = 0 (lled symbols). The solid line
indicates the quadratic t function used for extrapolation.
+
R
(z) g

RR
(z)
R
(z) g

RR
(z)
_
dz . (17)
This sum rule involves only the site-diagonal blocks of the auxiliary Green functions and its
reliable evaluation for perfect crystals requires only a few thousands of k-points in the irreducible
part of the BZ, i.e., accuracy usual in most selfconsistent LSDA calculations.
Another numerical problem is encountered in computations of the Curie temperature in the
RPA due to the singularity of the averaged function in Eq. (9) for [q[ 0. We have therefore
calculated T
RPA
C
using the expression
_
k
B
T
RPA
C
_
1
=
3
2
lim
z0
G
m
(z) , G
m
(z) =
1
N

q
_
z J
0
+J(q)
_
1
, (18)
where z is a complex energy variable and the quantity G
m
(z) is a magnon Green function
corresponding (up to the prefactor 4/M) to the magnon dispersion law, Eq. (6). The magnon
Green function was evaluated for energies z in the complex energy plane and its value for z = 0
was obtained using an analytical continuation technique [42].
4 Applications
4.1 Transition metals
Calculated magnon energy spectra E(q) for bcc Fe are presented in Fig. 4. Corresponding plots
of E(q) for fcc Co and Ni [14] exhibit parabolic, almost isotropic behavior for long wavelengths.
On the contrary, in bcc Fe we observe some anisotropy of E(q), i.e., E(q) increases faster along
the N direction and more slowly along the P direction. In agreement with Refs. [27, 43, 44]
we observe a local minima around the point H along H and HN directions in the range of
short wavelengths. They are indications of the so-called Kohn anomalies which are due to long-
range interactions mediated by the RKKY interactions similarly like Kohn-Migdal anomalies in
120
phonon spectra are due to long-range interactions mediated by Friedel oscillations. It should be
mentioned that minima in dispersion curve of bcc Fe appear only if the summation in Eq. (6)
is done over a suciently large number of shells, in the present case for more than 45 shells.
0
200
400
600
E
n
e
r
g
y

[
m
e
V
]
N P H N
Fe
Figure 4: Magnon dispersion law along high-symmetry lines in the Brillouin zone of bcc Fe
compared to experiment (lled circles: pure Fe at 10 K [45], empty squares: Fe(12% Si) at room
temperature [46]).
Present results for dispersion relations compare well with available experimental data of mea-
sured spin-wave spectra for Fe and Ni [45, 46, 47]. For low-lying part of spectra there is also a
good agreement of present results for dispersion relations with those of Refs. [27, 44] obtained
using the frozen-magnon approach. There are, however, dierences for a higher part of spectra,
in particular for the magnon bandwidth of bcc Fe which can be identied with the value of E(q)
evaluated at the high-symmetry point q = H in the bcc BZ. The origin of this disagreement is
unclear. We have carefully checked the convergence of the magnon dispersion laws E(q), see
Fig. 5, with the number of shells included in Eq. (6) and it was found to be weak for 50 70
shells and more, i.e., for the cut-o distance d
max
6a.
The results for spin stiness coecient D calculated for the three cubic ferromagnetic metals
are summarized in Tab. 1 together with available experimental data [48, 49, 50]. There is a
reasonable agreement between theory and experiment for bcc Fe and fcc Co but the theoretical
values of D are considerably overestimated for fcc Ni. It should be noted that measurements
refer to the hcp Co while the present calculations were performed for fcc Co. A similar agreement
between calculated and measured spin-wave stiness constants was obtained by Halilov et al.
[27] using the frozen-magnon approach. Our results are also in accordance with those obtained
by van Schilfgaarde and Antropov [44] who used the spin-spiral calculations to overcome the
problem of evaluation of D from Eq. (7). On the other hand, this problem was overlooked in
Refs. [11, 51, 52] so that a good agreement of D, calculated for a small number of coordination
shells, with experimental data seems to be fortuitous. Finally, the results of Brown et al. [53]
obtained by the layer KKR method in the frozen potential approximation are underestimated
for all metals and the best agreement is obtained for Ni.
Calculated values of Curie temperatures for both the MFA and RPA as well as corresponding
121
0
100
200
300
400
500
600
700
0 2 4 6 8 10
M
a
g
n
o
n

b
a
n
d
w
i
d
t
h

(
m
e
V
)
d /a
max
bcc Fe
Figure 5: The magnon bandwidth for bcc Fe as a function of the cut-o distance d
max
. The
bandwidth is identied with the magnon energy at the high-symmetry point H in the bcc
Brillouin zone.
Table 1: Calculated spin-wave stiness constants (D
th
) and Curie temperatures (T
MFA
C
and
T
RPA
C
) and their comparison with experimental values (D
exp
and T
C,exp
).
Metal D
th
[meV

A
2
] D
exp
[meV

A
2
] T
MFA
C
[K] T
RPA
C
[K] T
C,exp
[K]
Fe bcc 250 7 280, 330 1414 950 2 1044 1045
Co fcc 663 6 580, 510 1645 1311 4 1388 1398
Ni fcc 756 29 555, 422 397 350 2 624 631
experimental data are summarized in Tab. 1. MFA-values of Curie temperatures are overesti-
mated for Fe and Co, but underestimated for Ni in agreement with other calculations [27, 44].
On the other hand, the results obtained using the RPA are in a good agreement with experiment
for both fcc Co and bcc Fe, while the results for fcc Ni are even more underestimated. This is
in agreement with the fact mentioned in Section 2, namely that T
RPA
C
< T
MFA
C
. The present
results for Fe and Ni agree reasonably with results of Ref. [54] using the spin-uctuation theory
and an improved statistical treatment in the framework of the Onsager cavity-eld method.
In summary, we have found that calculated Curie temperatures and spin-wave stiness constants
agree well with experiment for Fe and Co, while less satisfactory results are obtained for Ni,
where the role of the Stoner excitations is much more important as compared to Fe and Co. In
addition, the adiabatic approximation is less justied for Ni, and, possibly, correlation eects
beyond the LSDA play the more important role for this ferromagnet.
4.2 Rare-earth metals
Rare-earth (RE) metals represent a class of systems where the concept of atomic-like local mo-
ments is well justied due to highly localized 4f orbitals. The standard LSDA, however, fails
to describe correctly their ground-state properties: the equilibrium lattice constants are signi-
122
cantly smaller than the experimental ones due to an overestimated 4f-contribution to cohesion,
and the ground-state magnetic structures in the LSDA are qualitatively wrong as well. In the
case of Gd in hcp structure, the antiferromagnetic (AFM) stacking of the (0001) atomic planes
was predicted [55] in contrast to the observed ferromagnetic (FM) state [56]. The most sophis-
ticated methods beyond the LSDA, which improve the situation, take explicitly into account
the on-site Coulomb interaction of the 4f electrons, like the LSDA+U scheme [57, 58] and the
self-interaction corrected (SIC) LSDA approach [59, 60]. Ground-state magnetic structures of
4f electron systems are often non-collinear and incommensurate with the underlying chemical
unit cell [56] which presents another complication for ab initio techniques.
We have treated two RE metals, namely, hcp Gd [17] and bcc Eu [18], in a simplied manner
taking the 4f states as a part of the atomic core (with the majority 4f level occupied by 7
electrons and the minority 4f level empty). The other valence orbitals were included in the
standard LSDA. This open-core approach was often employed in selfconsistent spin-polarized
calculations of RE-based systems during the last decade [60, 61, 62, 63] and it yielded the correct
FM structure of hcp Gd. The theoretical equilibrium Wigner-Seitz radii s (s
Gd
= 3.712 a.u.
with the experimental value of c/a = 1.597, s
Eu
= 4.190 a.u.) are only slightly smaller than
the experimental values (s
Gd
= 3.762 a.u., s
Eu
= 4.238 a.u.). Dierent spin congurations were
considered for both metals: FM, AFM, and the disordered local moment (DLM) state [21].
The theoretical equilibrium values of s are nearly insensitive to the spin structures and the FM
ground state of Gd exhibits a non-negligible energy separation from the AFM and DLM states
in contrast to the bcc Eu, where the DLM state is slightly more stable than the FM state.
0.0
0.1
0.2
1 2 3 4 5 6
J

(
m
R
y
)
d/a
Gd
0.0
0.1
0.2
1 2 3 4 5 6
J

(
m
R
y
)
d/a
Eu
Figure 6: Exchange interactions J
RR
for hcp Gd (left panel) and bcc Eu (right panel) as
functions of the distance [R R

[ = d. The crosses and squares in the left panel refer to pairs

of sites R, R

lying in even (AA) and odd (AB) hcp(0001) planes, respectively.

The exchange interactions in Gd and Eu, derived for the FM state and the theoretical equilibrium
Wigner-Seitz radius s, are shown in Fig. 6. Their distance-dependence is qualitatively similar
to the 3d transition metals, the magnitudes of the dominating nearest-neighbor interactions are,
however, smaller by a factor of ve, cf. Fig. 1. Moreover, as illustrated in Fig. 7 by calculating
the on-site exchange parameter J
0
as a function of the cut-o distance d
max
used in the real-
space sum in Eq. (8), there is a profound dierence between the two 4f metals concerning the
123
oscillating interactions of more distant atoms. In the hcp Gd, they are not strong enough to
destroy the FM spin structure, as indicated by the positive converged value of J
0
. Note that
the negative exchange interaction between the second Gd nearest neighbors is in qualitative
agreement with experiment [64]. On the other hand, the contribution of more distant sites to
J
0
is very important in the case of bcc Eu and it yields for the converged quantity a negligible
resulting value (J
0
= 0.03 mRy). Such a situation indicates an instability of the FM state
with respect to a more complicated spin structure. This feature agrees qualitatively with an
experimentally observed helical spin structure, the wave vector of which lies along the H
direction in the bcc BZ [65, 66].
1.0
1.5
2.0
2.5
3.0
3.5
1 2 3 4 5 6 7
Gd
J
0


(
m
R
y
)
-0.5
0.0
0.5
1.0
1.5
2.0
1 2 3 4 5 6 7 8 9 10
J
0


(
m
R
y
)
d
max
/a
Eu
Figure 7: The on-site exchange parameter J
0
for hcp Gd (top panel) and bcc Eu (bottom panel)
as a function of the cut-o distance d
max
used in the real-space summation in Eq. (8). The
horizontal lines mark exact values of J
0
obtained from the sum rule, Eq. (17).
Table 2: Calculated magnetic transition temperatures (T
MFA
and T
RPA
) and their comparison
with experimental values (T
exp
) for hcp Gd (Curie temperature) and bcc Eu (Neel temperature).
Calculations were performed with experimental values of lattice parameters.
Metal T
MFA
[K] T
RPA
[K] T
exp
[K]
Gd hcp 334 300 293
Eu bcc 151 111 91
Calculations of the magnon spectra and the Curie temperature for hcp Gd require a trivial
generalization of Eqs. (6, 9) to the case of two equivalent atoms in the hcp unit cell [67]. The
resulting Curie temperatures are given in Tab. 2 together with the experimental value [56] while a
comparison of the calculated magnon dispersion law with experiment [56] is presented in Fig. 8.
The theoretical magnon spectra included nite temperature of the experiment (T = 78 K)
124
which leads within the RPA to a simple rescaling of the magnon energies proportionally to the
temperature-dependent average magnetization [32]. The latter dependence was calculated in
the RPA from the classical EHH [67].
0
10
20
30
M K A
Gd
M
a
g
n
o
n

e
n
e
r
g
y

(
m
e
V
)
Figure 8: Magnon dispersion law along high-symmetry lines in the Brillouin zone of hcp Gd
calculated for T = 78 K (lines) and compared to experiment (lled circles - Ref. [56]).
The calculated magnon energies are higher than experimental. A recent theoretical study by
Halilov et al. [43, 62] revealed that this eect can be partly explained by assumed collinearity
between the localized 4f-moment and the valence part of the local moment. Inclusion of a
possible non-collinearity between the localized and itinerant moments leads to a softening of the
magnon energies, reducing them by a factor of 1.5 in the upper part of the spectrum. However,
the lower part of the spectrum that is more important for an RPA estimation of the Curie
temperature is less inuenced by the non-collinearity. On the other hand, the calculated Curie
temperatures both in the MFA and in the RPA agree very well with experiment (Tab. 2). This
degree of agreement proves that the present approach based on interatomic exchange interactions
represents a better starting point to RE magnetism than a theory based on intraatomic exchange
integrals formulated in Ref. [68]. The latter scheme provided values of the Curie temperature
for Gd in a wide interval between 172 K and 1002 K, depending on further approximations
employed.
Determination of the magnetic ground state of Eu from the EHH, Eq. (2), is a dicult task
in view of the highly-dimensional manifold of a priori possible states as well as a number of
qualitatively dierent spin structures encountered in RE-based systems [56]. Here we consider
only spin spirals specied by a single q-vector as
e
R
= (sin(q R), 0, cos(q R)) , (19)
since the spin structure observed for bcc Eu at low temperatures belongs to this class [65, 66].
The minimum of the Hamiltonian H
e
corresponds then to the maximum of the lattice Fourier
transform J(q), Eq. (6). A scan over the whole BZ reveals that the absolute maximum of J(q)
(for the theoretical equilibrium Wigner-Seitz radius s = 4.19 a.u.) is obtained for a vector q = Q
on the H line, namely, at Q = (1.69, 0, 0) a
1
, see Fig. 9. The magnitude of Q determines
the angle between magnetic moments in the neighboring (100) atomic layers. In the present
case, it is equal to = 48

. Similar values were obtained for the experimental value of the

125
Wigner-Seitz radius s
exp
= 4.238 a.u.: Q = (1.63, 0, 0) a
1
, = 47

. Both data sets are in

surprising agreement with experimental results which report the spin-spiral q-vector inside the
H line and the angle per layer equal to
exp
= 49

[65] and
exp
= 47.6 1.2

[66].
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
J
(
q
)

(
m
R
y
)
Eu
N P H N
Figure 9: The lattice Fourier transform J(q) of the exchange interactions in bcc Eu along
high-symmetry lines in the Brillouin zone.
The resulting maximum J(Q) can be used to get the Neel temperature in the MFA in complete
analogy to Eq. (8) [56]:
k
B
T
MFA
N
=
2
3
J(Q) , (20)
whereas the RPA leads to the following modication of Eq. (9) [32, 69]:
_
k
B
T
RPA
N
_
1
=
3
4
1
N

q
_
[J(Q) J(q)]
1
+ [W(q, Q)]
1
_
,
W(q, Q) = J(Q)
1
2
J(q +Q)
1
2
J(q Q) . (21)
Both theoretical values and the experimental Neel temperature are given in Tab. 2. The MFA-
value is substantially higher than experiment while the RPA reduces the theoretical value of T
N
signicantly so that a good agreement with experiment is obtained.
4.3 Substitutionally disordered alloys
The present real-space approach to exchange interactions can be generalized to substitutionally
disordered alloys either by using a supercell technique or by combining it with the coherent-
potential approximation (CPA). Both alternatives have their own merits and drawbacks. The
CPA takes properly into account the eects of nite lifetime of electronic states due to disorder
but it has diculties to include eects of varying local environments as well as of short-range
order (both chemical and magnetic) on electronic properties.
In the following, we sketch the modication of the expression for the exchange interactions,
Eq. (13), to a random alloy within the LMTO-CPA formalism [35, 36, 70]. We assume that the
lattice sites R are randomly occupied by alloy components Q = A, B, . . ., with concentrations
c
Q
R
. We neglect any correlations between occupations of dierent lattice sites and we neglect
local environment eects, i.e., the LSDA selfconsistent potentials inside R-th cell depend solely
on occupation of the site R by an atom Q = A, B, . . ..
126
The CPA-congurational average of the auxiliary Green function, Eq. (12), can be written as
g
RR
(z)) = g
RR
(z) =
_
[T(z) S]
1
_
RR

, (22)
where the spin index is omitted, S is the structure constant matrix and T(z) is a non-
random site-diagonal matrix of coherent potential functions T
R
(z) attached to individual lattice
sites which describe eective atoms forming an eective CPA medium. The coherent potential
functions satisfy a set of selfconsistency conditions (Soven equation) which guarantees that
average single-site scattering due to real atoms with respect to the eective medium vanishes.
The CPA leads also to conditional averages of individual blocks of the Green functions. The
site-diagonal block g
RR
(z) of the Green function averaged over all alloy congurations with site
R occupied by an atom Q is given by
g
Q
RR
(z) = g
RR
(z) f
Q
R
(z) =

f
Q
R
(z) g
RR
(z) , (23)
where the prefactors f
Q
R
(z) and

f
Q
R
(z) are dened as
f
Q
R
(z) =
_
1 +
_
P
Q
R
(z) T
R
(z)
_
g
RR
(z)
_
1
,

f
Q
R
(z) =
_
1 + g
RR
(z)
_
P
Q
R
(z) T
R
(z)
__
1
. (24)
Similarly, the site-o-diagonal block g
RR
(z) averaged over all alloy congurations with two sites
R ,= R

occupied respectively by atomic species Q and Q

is given by
g
QQ

RR

(z) =

f
Q
R
(z) g
RR
(z) f
Q

(z) . (25)
Derivation of the conditionally averaged pair exchange interaction between two sites R ,= R

occupied respectively by components Q and Q

can be performed similarly like in the case

without substitutional randomness by employing the magnetic force theorem [11] and the so-
called vertex-cancellation theorem [71, 72]. It leads to an expression

J
QQ

RR

=
1
8i
_
C
tr
L
_

Q
R
(z) g
QQ

,
RR

(z)
Q

(z) g
Q

Q,
R

R
(z)
_
dz ,

Q
R
(z) = P
Q,
R
(z) P
Q,
R
(z) , (26)
which is fully analogous to Eq. (13). The conditional average of the on-site exchange interaction,
Eq. (17), yields a formula

J
0,Q
R
=
1
8i
_
C
tr
L
_

Q
R
(z)
_
g
Q,
RR
(z) g
Q,
RR
(z)
_
+
Q
R
(z) g
Q,
RR
(z)
Q
R
(z) g
Q,
RR
(z)
_
dz . (27)
It should be noted, however, that the sum rule for the averaged pair and on-site interactions,

J
0,Q
R
=

R

J
QQ

RR

c
Q

, (28)
which can be easily obtained from the corresponding sum rule, Eq. (17), valid for any congu-
ration of the alloy, is not exactly satised by the expressions (26) and (27). According to our
127
experience, the two sides of Eq. (28) deviate up to 15% for typical binary transition-metal alloys
(FeV, FeAl). This violation of an important sum rule indicates that vertex corrections must
be taken into account in averaging exchange interactions in random alloys. On the other hand,
the small relative dierence of both sides of the sum rule (28) proves that the role of vertex
corrections for exchange interactions is less signicant than in transport properties, as argued
in Ref. [73].
Let us now consider the case of two isolated impurities in a non-magnetic host. The exchange
interaction between impurity sites R ,= R

can be calculated exactly and compared to the low-

concentration limit of the CPA expression, Eq. (26). The latter case corresponds to a binary
alloy with c
A
R
0 and c
B
R
1 for all lattice sites, with the coherent potential functions
T

R
(z) P
0
R
(z) where P
0
R
(z) = P
B
R
(z) are the non-spin-polarized host potential functions, and
with the average Green function substituted by that of the non-random host, g

RR
(z) g
0
RR
(z).
Direct calculation of the exchange interaction for two impurities embedded in a crystal gives a
result (the so-called two-potential formula [74]) which diers from the low-concentration limit
of the CPA expression by matrix quantities X

RR
(z) dened in terms of single-site t-matrices
of impurities

R
(z) as
X

RR
(z) =
_
1 g
0
RR
(z)

R
(z) g
0
R

R
(z)

R
(z)
_
1
,

R
(z) =
_
P
A,
R
(z) P
0
R
(z)
_ _
1 + g
0
RR
(z)
_
P
A,
R
(z) P
0
R
(z)
__
1
. (29)
The quantities X

RR
(z) which enter the impurity-impurity interaction as multiplicative pref-
actors at the site-o-diagonal blocks g
0
RR
(z) of the host Green function describe multiple
scatterings of electrons between the two impurity sites; their absence in the exchange interac-
tion derived from the low-concentration limit of Eq. (26) reects a systematic neglect of such
multiple-scattering processes in single-site theories like the CPA.
-0.5
0.0
0.5
1.0
1.5
2.0
1 2 3 4 5 6 7 8 9 10
J


(
m
R
y
)
d/a
Fe - 20% Al
-1.0
-0.5
0.0
0.5
1.0
1.5
1 2 3 4 5 6 7 8 9 10
(
d
/
a
)
3

J


(
m
R
y
)
d/a
Fe - 20% Al
Figure 10: Exchange interactions

J
FeFe
RR
for a random bcc Fe
0.8
Al
0.2
alloy as a function of the
distance [RR

[ = d without (left panel) and with (right panel) a prefactor d

3
.
An example of pair exchange interactions is shown in Fig. 10 for Fe-Fe pairs in a disordered bcc
Fe
0.8
Al
0.2
alloy. The interactions are qualitatively similar to those in pure bcc Fe (Fig. 1), but
128
a more careful analysis of the long-distance behavior reveals an exponentially damped RKKY-
like oscillations (Fig. 10, right panel). This feature can be explained by damping of electron
states due to the alloy disorder which leads to exponential decay of site-o-diagonal blocks of
the averaged Green functions g

RR
(z) with increasing distance [RR

[. It should be mentioned
that this exponential damping refers only to averaged exchange interactions in contrast to those
in each alloy conguration which exhibit a much slower decay for large interatomic separations
(see Ref. [73] and references therein).
Calculation of magnon spectra in disordered alloys represents a non-trivial task since the cor-
responding equation of motion for the two-time Green function for spin operators, obtained
from the standard decoupling procedure for higher-order Green functions [32], contains a more
complicated type of disorder than purely diagonal disorder. The magnons (and also phonons)
in random alloys are featured by simultaneous presence of diagonal, o-diagonal and environ-
mental disorder; the latter is closely related to the Goldstone theorem for these excitations.
An extension of the CPA to this case has been studied since early 1970s. Two most recent
approaches are based on a cumulant expansion [75] and on an augmented-space formalism [76];
the former scheme is combined with the RPA and provides a value for the Curie temperature.
Both formulations are rather complicated which allowed to perform numerical calculations for
environmental disorder limited to nearest neighbors only, but they seem promising for future
studies with true long-range interactions.
The Curie temperature of a random alloy in the MFA can be obtained in a way similar to that
leading to Eq. (8). Let us restrict ourselves to the case of a homogeneous random alloy (with all
lattice sites equivalent). In analogy to previous on-site exchange parameters, Eqs. (8, 27), one
can introduce quantities

J
0,QQ

=

R

J
QQ

RR

, /
QQ

=

J
0,QQ

c
Q

, (30)
where /
QQ

are eective exchange parameters among magnetic moments of the alloy con-
stituents. The Curie temperature is then equal to
k
B
T
MFA
C
=
2
3

max
, (31)
where
max
is the maximal eigenvalue of the matrix /
QQ

. This type of estimation has been

used for diluted magnetic semiconductors as described in Section 4.4.
4.4 Diluted magnetic semiconductors
Diluted magnetic semiconductors (DMS) represent a new class of materials with potential tech-
nological applications in spintronics. They have recently attracted much interest because of the
hole-mediated ferromagnetism [77, 78]. Curie temperatures higher than the room temperature
are desirable for practical applications, whereas the currently prepared samples exhibit the T
C
s
only slightly above 100 K [78, 79]. The most frequently studied DMS is a III-V-based compound
(Ga
1x
Mn
x
)As in the zinc-blende structure with Mn-concentration in the range 0 < x < 0.1.
Since Mn atoms are in a high-spin state in these systems, the above described formalism is
well suited for reliable quantitative investigations of the exchange interactions and the Curie
temperatures.
129
0
1
2
0 1 2 3
M
n
-
M
n


e
x
c
h
a
n
g
e


i
n
t
e
r
a
c
t
i
o
n


(
m
R
y
)
d/a
(Ga,Mn,As)As
y = 0.0
y = 0.01
-4
-2
0
2
4
0 0.05 0.1
1
s
t
-
n
n


J


(
m
R
y
)
Mn-concentration x
(Ga,Mn,As)As
y = 0.0
y = 0.005
y = 0.01
Figure 11: The Mn-Mn exchange interactions in (Ga
0.95y
Mn
0.05
As
y
)As as a function of
the Mn-Mn distance d (left panel) and the rst nearest-neighbor Mn-Mn interaction in
(Ga
1xy
Mn
x
As
y
)As (right panel).
The (Ga,Mn)As compound is a substitutionally disordered system with Mn atoms substituting
Ga atoms on the cation sublattice. Application of the TB-LMTO-CPA formalism to this sys-
tem employs so-called empty spheres located at interstitial positions of GaAs semiconductor for
matters of space lling, so that the zinc-blende structure is described in terms of four fcc sublat-
tices with substitutional disorder only on the cation sublattice. The pair exchange interactions
between Mn atoms

J
MnMn
RR
in the (Ga
1x
Mn
x
)As alloy with x = 0.05 are shown in Fig. 11 (left
panel); interactions between the other components are much smaller and negligible concerning
their possible inuence on magnetic properties. The rst nearest-neighbor interaction is positive
and bigger than the (mostly positive) interactions between more distant Mn atoms. Analysis
of the behavior of

J
MnMn
RR
for large interatomic distances reveals exponentially damped RKKY-
like oscillations which have two origins: the eect of alloying which introduces an exponential
damping in the site-o-diagonal blocks of the averaged Green functions for both spin channels
(see Section 4.3), and an additional exponential damping due to a half-metallic character of the
system [77], i.e., the alloy Fermi energy lies in a band gap of the minority band (see Section 3.3).
The calculated Curie temperature in MFA for the (Ga
0.95
Mn
0.05
)As alloy is around 300 K, i.e.,
substantially higher than the experimental values [78, 79]. A reason for this discrepancy can be
found in structural imperfections of the compound which lead to reduction of the number of holes
in the valence band. The most probable candidates for such lattice defects are native defects,
such as As-antisite atoms [77] and Mn-interstitial atoms [80]. In the following, we demonstrate
the inuence of the former on the exchange interactions of (Ga,Mn)As compounds [18, 20].
The combined eect of Mn-impurities and As-antisites can be simulated within the CPA using
an alloy (Ga
1xy
Mn
x
As
y
)As with y denoting the As-antisite concentration. The inuence of
As-antisites on the Mn-Mn exchange interactions is shown in Fig. 11 (left panel): the positive
values of

J
MnMn
RR
are reduced; the most dramatic reduction is found for the dominating coupling
between the nearest-neighbors. The dependence of the nearest-neighbor Mn-Mn interaction on x
130
and y is shown in Fig. 11 (right panel). For a xed Mn-concentration x, the interaction decreases
monotonously with increasing content of As-antisites y, ending nally at negative values. This
change of sign correlates nicely with a predicted instability of the ferromagnetic state with
respect to formation of a state featured by disordered directions of the Mn-moments [16, 81].
0
100
200
300
0 0.05 0.1 0.15
C
u
r
i
e


t
e
m
p
e
r
a
t
u
r
e


(
K
)
Mn-concentration x
y=0.0
y=0.01
exp.
Figure 12: Curie temperatures of (Ga
1xy
Mn
x
As
y
)As: calculated (full symbols) and experi-
mental [78] (open squares).
The Curie temperatures were estimated in the MFA as described in Sections 2 and 4.3. However,
in view of the much bigger Mn-Mn interactions as compared to interactions between other
constituent atoms, the Curie temperature comes out equal to
k
B
T
MFA
C
=
2
3
x

J
MnMn
RR
, (32)
where the lattice sites R, R

are conned to the cation fcc sublattice and x denotes the Mn-
concentration. The resulting Curie temperatures are summarized in Fig. 12 [16]. The T
MFA
C
for
a xed x is monotonously decreasing with increasing As-antisite concentration y, in analogy to
the y-dependence of the rst nearest-neighbor Mn-Mn interaction (Fig. 11, right panel). The
T
C
for a xed y exhibits a non-monotonous dependence on the Mn-content x reaching a at
maximum for x > 0.1. The latter behavior results from an interplay of two eects: an increase
of T
MFA
C
with increasing x, Eq. (32), and the non-trivial dependence of the rst nearest-neighbor
Mn-Mn interaction as a function of (x, y), see Fig. 11 (right panel). Note, however, that the
next-neighbor exchange couplings also contribute signicantly to the Curie temperature, see
Eq. (32). A detailed comparison of calculated and measured concentration dependences of
the Curie temperature indicates a correlation between the two concentrations x and y in real
samples, namely, an increase of the As-antisite concentration with increasing Mn-content [16].
However, this possible explanation of the measured Curie temperatures does not rule out other
lattice defects in the system.
Curie temperatures of DMSs have recently been calculated from rst principles using alternative
approaches [20, 82, 83]. The simplest estimation is based on the total-energy dierence E
between the DLM state and the ferromagnetic state [16, 82]. The quantity E can be obtained
from selfconsistent calculations using the CPA and it can be within the EHH and the MFA
131
-80
-60
-40
-20
0
20
40
60
J
(
q
)

(
m
R
y
)
L X W K
Mn-Mn
Figure 13: The lattice Fourier transform of the Mn-Mn exchange interaction for
(Ga
0.95y
Mn
0.05
As
y
)As along high-symmetry lines of the bcc Brillouin zone: y = 0.0 (full line),
y = 0.01 (long dashes), y = 0.015 (short dashes), y = 0.02 (dotted line), y = 0.025 (dashed-
dotted line).
identied with the lattice sum in Eq. (32) multiplied by x
2
, which yields an expression
k
B

T
MFA
C
=
2 E
3 x
. (33)
According to our experience for (Ga,Mn)As alloys [16], the values of

T
MFA
C
are higher by 10 to
15% as compared to the values of T
MFA
C
from Eq. (32).
A combination of the frozen-magnon and supercell approaches was used to study Curie temper-
atures in (Ga,Mn)As (without structural defects) in the MFA and the RPA [83]. It yielded a
non-monotonous dependence of the T
C
on the Mn-concentration, very similar to that depicted
in Fig. 12 (results for y = 0). The RPA values were about 20% smaller than the MFA values;
the latter compare well with the present results. It should be noted, however, that the supercell
approach was limited to a few special Mn-concentrations (x = 0.03125, 0.0625, 0.125, 0.25) and
that the rst nearest-neighbor Mn-Mn interactions could not be determined due to the special
atomic order of the supercells.
Probably the most reliable way of obtaining the Curie temperature from parameters of the EHH
is the Monte Carlo simulation. A recent investigation for (Ga,Mn)As alloys proved that the
Monte Carlo results yield Curie temperatures only slightly smaller (less than 10%) as compared
to the MFA while an RPA estimation of the Curie temperature was found between the MFA and
the Monte Carlo values [20]. This success of MFA can be explained by a few special features
of the Mn-Mn exchange interactions: they are essentially ferromagnetic, not oscillating, and
decaying exponentially with increasing distance, see Fig. 11 (left panel). Their lattice Fourier
transforms, shown in Fig. 13, become rather dispersionless, except for a small region around
the point, which leads to the small dierences among the Curie temperatures evaluated in
dierent ways [20].
Group-IV DMSs, like Mn-doped Ge, have been studied only recently [84, 85], but their Curie
temperatures are very similar to those of III-V DMSs. The main dierence between the two
classes of DMSs lies in positions occupied by Mn atoms: they are located on one fcc sublattice
of the zinc-blende structure in the (Ga,Mn)As compound (without native defects) but they
132
-2
-1
0
1
2
3
0 1 2 3
J
M
n
,
M
n

(
m
R
y
)
d/a
Ge
0.95
Mn
0.05
same
different
fcc sublattices
-4
-2
0
2
4
6
8
0 0.02 0.04 0.06 0.08
1
s
t


a
n
d


2
n
d


N
N


J
M
n
,
M
n

(
m
R
y
)
Mn-concentration x
Ge
1-x
Mn
x
2nd
1st
Figure 14: The Mn-Mn exchange interactions in Ge
0.95
Mn
0.05
as a function of the Mn-Mn
distance d (left panel) and the rst and the second nearest-neighbor Mn-Mn interactions in
Ge
1x
Mn
x
(right panel). The pair interactions are divided according to positions of the two Mn
atoms on two fcc Ge-sublattices.
occupy lattice sites of two fcc sublattices of the diamond structure in the (Ge,Mn) case. The
latter system thus contains Mn-Mn pairs with a short distance which is known to support
antiferromagnetic exchange coupling of Mn-moments.
-2
-1
0
1
2
3
4
5
6
1 3 5 7 9 11 13 15 17
(
d
/
a
)
3

J
M
n
,
M
n


(
m
R
y
)
(d/a) along bond-direction
Ge-Mn
x = 0.005
x = 0.025
Figure 15: The Mn-Mn exchange interactions multiplied by d
3
in Ge
1x
Mn
x
as a function of
the Mn-Mn distance d for Mn-Mn pairs along the bond direction.
The calculated Mn-Mn exchange interactions in (Ge,Mn) alloys are shown in Fig. 14. It can
be seen that the interactions exhibit strong concentration dependence, as illustrated in right
panel of Fig. 14 for the rst and second nearest-neighbor interactions: for 6% of Mn we nd
antiferromagnetic coupling between Mn neighbors on dierent fcc sublattices in a qualitative
agreement with results of supercell calculations in Ref. [84]. The negative rst nearest-neighbor
interaction appears for alloys with more than 3% Mn, whereas the next-neighbor interactions
133
are essentially ferromagnetic (Fig. 14, left panel). Note that the concentration dependence of
the second nearest-neighbor interaction (Fig. 14, right panel) is very similar to that between the
rst nearest Mn-Mn neighbors in the Ga
1x
Mn
x
As alloy (Fig. 11, right panel).
The asymptotic behavior of

J
MnMn
RR
is presented in Fig. 15 for Mn-Mn pairs along the bond
direction, i.e., along a zig-zag line following the [110] direction. Besides the exponential damping
of the RKKY-type oscillations, discussed above, one can see a pronounced change of the periods
of oscillations with Mn-concentration. This property agrees fully with the RKKY picture which
leads to oscillation periods inversely proportional to the characteristic size of the hole of the
Fermi surface. It can be also seen that the ferromagnetic character of couplings for x = 0.025
is preserved up to a distance of about 4a (a is the fcc lattice constant) which is bigger than the
average distance of about 3.4a between two Mn-impurities.
0
50
100
150
200
0 0.01 0.02 0.03
C
u
r
i
e

t
e
m
p
e
r
a
t
u
r
e

(
K
)
Mn-concentration x
Ge-Mn
MFA
exp
Figure 16: Curie temperatures in Ge
1x
Mn
x
: calculated in MFA and experimental [84].
Curie temperatures in the MFA for Ge
1x
Mn
x
alloys were calculated according to Eq. (32) with
the lattice sum extending over two fcc sublattices occupied by Mn atoms. The concentration
dependence is shown in Fig. 16 together with available experimental data [84]. The calculated
Curie temperatures increase with Mn-content up to x 0.03 where a saturated behavior appears
as a consequence of the antiferromagnetic coupling between the rst Mn-Mn neighbors (Fig. 14).
The measured Curie temperatures increase with Mn-content as well; the dierence between the
theoretical and experimental data is probably due to Mn-interstitials present in real samples.
Their inclusion into theoretical models is under way.
4.5 Two-dimensional ferromagnets
Magnetism of epitaxial ultra-thin transition-metal lms on non-magnetic noble- or transition-
metal substrates has been studied intensively during the last two decades. Main dierences with
respect to bulk magnetism lie both in ground-state properties, where non-zero local moments
can appear also for other elements besides the ve 3d transition metals (Cr, Mn, Fe, Co, Ni)
[86], and in nite-temperature behavior, where the reduced dimensionality leads to a dierent
universality class as compared to the bulk. In the limit of one-monolayer thickness of the lm,
one can realize a true two-dimensional magnet on a non-magnetic substrate.
The above formalism can easily be generalized to the two-dimensional case since the basic
expression for the pair exchange interactions, Eq. (13), is formulated in the real space. The
134
magnetic properties resulting from a two-dimensional EHH can be obtained in a similar way
like in the bulk case, see Eqs. (6, 7, 8, 9), with the reciprocal-space vector q replaced by a
two-dimensional vector q

in the surface Brillouin zone (SBZ) and with the real-space sums
restricted to lattice sites R, R

of the magnetic lm. The site-o-diagonal blocks g

RR
(z) of the
Green function in Eq. (13) are determined using the surface Green function technique [35, 36],
while the denition of
R
(z) remains unchanged. The magnon energies are given by
E(q

) =
4
M
_
J(0

) J(q

)
_
+ , J(q

) =

R
J
0R
exp(iq

R) , (34)
where is a magnetic anisotropy energy which is a consequence of relativistic eects (spin-orbit
interaction, magnetostatic dipole-dipole interaction). The Curie temperature in the MFA is
given by Eq. (8) while the RPA leads to an expression
_
k
B
T
RPA
C
_
1
=
6
M
1
N

1
E(q

)
, (35)
where N

is the number of q

-vectors used in the SBZ-average.

-1
0
1
1 2 3 4 5 6 7
(
d
/
a
)
2

J


(
m
R
y
)
d/a
Co
Figure 17: Exchange interactions J
RR
multiplied by a prefactor d
2
for Co-moments in a Co-
overlayer on an fcc Cu(001) substrate as a function of the distance [RR

[ = d.
The calculated pair exchange interactions J
RR
in a Co-monolayer on an fcc Cu(001) substrate
are shown in Fig. 17. The rst nearest-neighbor interaction dominates and the next-neighbor
interactions exhibit an RKKY-like oscillatory behavior with an envelope decaying proportionally
to [R R

[
2
, in contrast to the bulk decay proportional to [R R

[
3
. Note, however, that
the present case is not strictly two-dimensional due to the indirect exchange interactions of two
Co-overlayer atoms via the Cu-substrate, which becomes weakly polarized in the atomic layers
adjacent to the overlayer.
The indirect interaction between the magnetic atoms, which is mediated by the non-magnetic
atoms, has important consequences for magnetic properties of magnetic lms placed on a non-
magnetic substrate and covered by a non-magnetic cap-layer of a nite thickness. As reported
in a recent experiment [87], the Curie temperature of fcc(001)-Fe ultrathin lms on a Cu(001)
substrate varies in a non-monotonous manner as a function of the Cu cap-layer thickness. Such
a behavior clearly cannot be explained within a localized picture of magnetism.
Motivated by this nding we performed a systematic study of Fe- and Co-monolayers on an fcc
Cu(001) substrate capped by another Cu-layer of varying thickness [13, 15]. Figure 18 presents
the magnon spectra in two limiting cases, namely, for an uncovered Fe-overlayer on Cu(001)
135
0
100
200
300
400
500
E
n
e
r
g
y

[
m
e
V
]
M X M
Fe
DOS
Figure 18: Magnon dispersion laws (left panel) and corresponding densities of states (right panel)
for an Fe-layer embedded in fcc Cu (full lines) and an Fe-overlayer on fcc Cu(001) (dashed lines).
We have set here = 0 in Eq. (34).
and for an Fe-monolayer embedded in bulk Cu, and gure 19 shows the full dependence of
the magnetic moments and the rst nearest-neighbor exchange interactions on the cap-layer
thickness. The magnon spectra and the magnon densities of states exhibit all typical features
of two-dimensional bands with the nearest-neighbor interactions which are here only slightly
modied by non-vanishing interactions in next shells. The magnetic moments drop substantially
on capping while their sensitivity to increasing cap-layer thickness is rather small. On the other
hand, the behavior of the nearest-neighbor exchange interaction is more complicated and it
reects interference eects in the Cu-cap layer. The oscillations visible in right panel of Fig. 19
are due to quantum-well states in the Cu-cap layer formed between the vacuum and the magnetic
layer which, in turn, inuence properties of the magnetic layer. Note that the values of the
nearest-neighbor exchange interaction are signicantly enhanced (roughly by a factor 2 or more)
as compared with their bulk counterparts (cf. Fig. 1).
Calculations of the Curie temperatures of the two-dimensional ferromagnets represent a more
dicult task than in the bulk case. The MFA Curie temperatures of the monolayers are typically
of the same order of magnitude as the corresponding bulk temperatures [13] due to the fact
that the reduced coordination is approximately compensated by the increase of the exchange
interactions. This observation is in a strong disagreement with experimental data which yield
the Curie temperatures of the order 150 200 K. This failure is due to the fact that the MFA
violates the Mermin-Wagner theorem [88] due to the neglect of collective transverse uctuations
(spin-waves) and it is thus inappropriate for two-dimensional systems.
Application of the RPA to the Curie temperature of a two-dimensional isotropic EHH, Eq. (35)
with = 0, yields a vanishing T
RPA
C
in agreement with the Mermin-Wagner theorem. Finite
values of T
RPA
C
require non-zero values of the magnetic anisotropy energy which is taken here
as an adjustable parameter. This is not a serious problem as the RPA Curie temperature has
136
1.5
1.6
1.7
1.8
2.6
2.7
2.8
2.9
0 5 10 15
M
a
g
n
e
t
i
c

m
o
m
e
n
t

(
B
o
h
r

m
a
g
n
e
t
o
n
s
)
Cap-layer thickness (MLs)
Fe
Co
1.9
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
0 5 10 15
1
s
t
-
N
N

e
x
c
h
a
n
g
e

i
n
t
e
r
a
c
t
i
o
n

(
m
R
y
)
Cap-layer thickness (MLs)
Fe
Co
Figure 19: The magnetic moments (left panel) and the rst nearest-neighbor exchange interac-
tions (right panel) of the Fe- and Co-monolayers on fcc Cu(001) as a function of the cap-layer
thickness. The dashed lines represent the embedded layer limit (innite cap thickness) while the
limit of zero cap thickness corresponds to the uncovered overlayer.
only a weak logarithmic dependence upon [89], and it is thus sucient to know the order of
magnitude of . The latter is typically of the order of the dipolar energy 2(M
B
)
2
/V , where
V is the atomic volume. In calculations we used
Co
= 0.052 mRy and
Fe
= 0.140 mRy.
The calculated RPA Curie temperatures are shown in Fig. 20 (left panel). They are strongly
reduced as compared to the corresponding bulk values thereby improving on the MFA results.
Nevertheless, they are still too large as compared to experiment. It is unclear whether this is due
to some inaccuracy of the theory or to some imperfections of the samples used in experiments.
On the other hand, such important experimental facts as the strong inuence of the metallic
coverage on the Curie temperature [87] are well explained by the present theory. The uctuations
of the Curie temperature are of order of 50 70 K, in a reasonable agreement with experiment.
A more detailed analysis of the data reveals that the oscillations of the RPA Curie temperatures
follow rather closely the behavior of the spin-stiness constants, see Fig. 20. The similarity
of both trends is due to the fact that the T
RPA
C
for a two-dimensional system is determined
predominantly by low-energy magnons.
It should be noted that an analogous oscillatory behavior of the Curie temperature as a function
of the non-magnetic spacer thickness has also been observed for fcc(001)-Co/Cu/Ni trilayers
[90]. The latter system has been investigated theoretically in terms of the on-site exchange
parameters J
0
R
[91].
137
200
300
400
500
600
700
0 5 10 15
C
u
r
i
e

t
e
m
p
e
r
a
t
u
r
e

(
K
)
Cap-layer thickness (MLs)
Fe
Co
300
400
500
300
400
500
600
0 5 10 15
S
t
i
f
f
n
e
s
s

c
o
n
s
t
a
n
t

(
m
e
V

A

)
2
o
.
Cap-layer thickness (MLs)
Fe
Co
Figure 20: The Curie temperatures (left panel) and the spin-stiness constants (right panel)
of the Fe- and Co-monolayers on fcc Cu(001) as a function of the cap-layer thickness. The
dashed lines represent the embedded layer limit (innite cap thickness) while the limit of zero
cap thickness corresponds to the uncovered overlayer case.
4.6 Surfaces of ferromagnets
Reduced coordination at surfaces of transition-metal ferromagnets leads to an enhancement
of surface magnetic moments over their bulk values [86]. For the ferromagnetic hcp Gd, an
enhancement of its Curie temperature at (0001) surface over the bulk value was observed [92].
Theoretical explanation of the latter fact was provided by total-energy calculations using an
LSDA+U approach [93]. An important role was ascribed to a small inward relaxation of the
top surface layer. However, more recent works have thrown serious doubts on these conclusions,
both on side of experiment [94] and theory [63].
We have recently performed calculations for low-index surfaces of bcc Fe, hcp Co, and hcp Gd [19]
focused on layer-resolved local quantities like the magnetic moments and the on-site exchange
parameters J
0
R
, Eq. (17). Note, however, that for inhomogeneous systems like surfaces, a direct
relation between the Curie temperatures and the on-site exchange parameters J
0
R
cannot be
given. Hence, the latter quantities reect merely the strength of the exchange interaction and
its spatial variations in layered systems [91].
Figure 21 presents the results for Fe- and Co-surfaces. It is seen that the well-known surface
enhancement of the moments is accompanied by a more complicated layer-dependence of the
on-site exchange parameters exhibiting a minimum in the top surface layer and a maximum
in the rst subsurface layer. A qualitative explanation follows from Eqs. (13, 17) which show
that J
0
R
reects the exchange splitting on the R-th site as well as the splittings and number
of its neighbors. Hence, the reduction of J
0
R
in the top surface layer is due to the reduced
coordination, whereas the maximum in the rst subsurface position is due to the full (bulk-like)
138
1.5
2.0
2.5
3.0
0 1 2 3 4 5 6
M
a
g
n
e
t
i
c

m
o
m
e
n
t

(

B
)
Layer
Fe(001)
Fe(110)
Co(0001)
6
10
14
18
0 1 2 3 4 5 6
O
n
-
s
i
t
e

e
x
c
h
a
n
g
e

p
a
r
a
m
e
t
e
r

(
m
R
y
)
Layer
Co(0001)
Fe(001)
Fe(110)
Figure 21: Layer-resolved magnetic moments (left panel) and on-site exchange parameters J
0
R
(right panel) at surfaces of bcc Fe and hcp Co. The layer numbering starts from the top surface
layer, denoted by 0.
coordination of these sites and the enhanced surface local moments, see Fig. 21. Note that the
layer-dependence of the on-site exchange parameters and its explanation are analogous to the
case of hyperne magnetic elds at the nuclei of iron atoms [35, 95].
The Gd(0001) surface was treated in the open-core approach mentioned in Section 4.2. Two
models of the surface structure were used: with lattice sites occupying the ideal truncated bulk
positions (unrelaxed structure) and with a 3% contraction of the interlayer separation between
the two topmost atomic layers (inward relaxation). The magnitude of the contraction was set
according to LEED measurements [96] and previous full-potential calculations [63]. The layer-
resolved quantities are presented in Figure 22. The layer-resolved magnetic moments exhibit
a small surface enhancement followed by Friedel-like oscillations around the bulk value. These
oscillations can be resolved also in the layer-dependence of the on-site exchange parameters J
0
R
which, however, start with reduced values in the top surface layer due to the reduced coordina-
tion, as discussed for Fe- and Co-surfaces. The maximum of the on-site exchange parameters
is found in the second subsurface layer, in contrast to the transition-metal surfaces, which can
be explained by the reduced Gd-moments in the rst subsurface layer. The surface relaxation
does not modify investigated layer-dependences substantially: it leads to a small reduction of
the local moments and the on-site exchange parameters in the rst two top surface layers and a
tiny enhancement in the second subsurface layer as compared to the ideal surface.
One can conclude that the surface enhancement of the local magnetic moments of the three
ferromagnetic metals is not accompanied by an analogous trend of on-site exchange parameters
which might be an indication of a surface-induced enhancement of Curie temperatures. However,
a calculation of the pair exchange interactions and an improved treatment of the EHH beyond
the MFA remain important tasks for future.
139
7.4
7.5
7.6
7.7
7.8
7.9
8.0
0 1 2 3 4 5 6
M
a
g
n
e
t
i
c

m
o
m
e
n
t

(

B
)
Layer
Gd(0001)
ideal
relaxed
2.0
3.0
4.0
0 1 2 3 4 5 6
O
n
-
s
i
t
e

e
x
c
h
a
n
g
e

p
a
r
a
m
e
t
e
r

(
m
R
y
)
Layer
Gd(0001)
ideal
relaxed
Figure 22: Layer-resolved magnetic moments (left panel) and on-site exchange parameters J
0
R
(right panel) at the (0001) surface of hcp Gd as calculated with the lattice sites in the ideal
truncated bulk positions and with the top surface layer relaxed towards the bulk. The layer
numbering starts from the top surface layer, denoted by 0.
5 Discussion and outlook
Exchange interactions and related quantities can be extracted from selfconsistent electronic
structure calculations in a number of dierent ways; each of them has its own advantages
and disadvantages. The simplest approach lies in obtaining the exchange interactions from
total-energy dierences calculated directly for various (usually collinear) spin congurations
[63, 97, 98]. The usefulness of such schemes is, however, rather limited mainly due to a nite
number of congurations used for the mapping.
The real-space method presented here allows to get a set of pair interactions even for very distant
atoms only from a single selfconsistent calculation. It has therefore been used independently
in a number of papers for very dierent systems [51, 99, 100, 101]. A widely used alternative
approach, the so-called frozen-magnon approach, relies on constrained DFT calculations for true
spin spirals and subsequent derivation of the coupling constants from the total energies of the
spirals [27, 43, 44]. The frozen-magnon approach can be implemented with the magnetic force
theorem [83] or can be formulated in terms of torques instead of total energies [102]. The real-
space and frozen-magnon approaches are formally equivalent to each other. The quantities that
are directly calculated (the pair exchange interactions J
RR
in the former case, the magnon
energies E(q)s in the latter) are related to each other by a lattice Fourier transformation,
Eq. (6). For this reason, their advantages and disadvantages refer mainly to their computational
eciency. For calculations of spin-wave dispersion curves (for a moderate number of q-points)
and of the spin-wave stiness D, the frozen-magnon approach is superior, for it does not require
to perform the Fourier transformation and the delicate analysis explained in Section 3.3. The
real-space approach seems to be more ecient for calculations of the Curie temperature, where
the MFA estimation can be obtained from Eq. (8) and the sum rule (17), i.e., from a single
real-space calculation, whereas a BZ-average of the magnon energies E(q) is required in the
140
frozen-magnon approach. The real-space technique is better also when a ne scan over the full
BZ is needed like, e.g., in RPA calculations of the Curie temperature, Eq. (9), or in searching
for instabilities of the ferromagnetic state (see Section 4.2), since the set of calculated J
RR
for
typically 200 shells provides a fast and accurate parametrization of J(q), which considerably
reduces the computational eort.
The central idea of mapping the innitesimal changes of single-particle energies onto an eec-
tive classical Hamiltonian for localized spins has been used also for a quantitative description of
eects beyond the bilinear isotropic exchange interaction in Eq. (2). Higher terms in expansion
of the single-particle energies with respect to rotation angles give rise to biquadratic exchange
interaction, important especially in metallic multilayers [103]. Inclusion of relativistic eects,
leading to anisotropic exchange interaction and Dzialoshinskii-Moriya interaction, has been dis-
cussed for bulk systems [52, 104] and thin lms [105]. In both cases, the gap in the spin-wave
spectrum for zero wavevector can be calculated with accuracy better than used in Section 4.5
for quantity in Eq. (34). The magnetic force theorem and expressions for eective interaction
parameters have recently been formulated also for highly correlated systems treated beyond the
LSDA [106].
It should be noted, however, that certain aspects of nite-temperature itinerant magnetism
cannot be reproduced by eective Hamiltonians with one unit vector per atomic site only. An
ab initio study based on a model Hamiltonian with bilinear, biquadratic and bicubic terms
depending on the magnitude of local magnetic moments was formulated in Ref. [107]. The
parametrization of the model was obtained from selfconsistent calculations for a number of spin-
spiral states. The calculated Curie temperature of bcc Fe agrees well with experiment while the
Curie temperatures of fcc Co and fcc Ni were underestimated by about 20 25%.
Another ab initio approach based on eective Hamiltonians acting on more than one magnetic
vector per atom was worked out in Refs. [43, 62]. It is based on a well-dened spatial separability
of the sp- and d-components of the spin density in late 3d transition metals leading thus to
two spin vectors per atom. Results of this approach for the magnon spectrum of fcc Ni seem
to describe qualitatively well ner details in the experimental data around 100 150 meV
(sometimes denoted as an optical magnon), which are manifestation of non-adiabatic eects
[43]. In the case of heavy RE-metals (Gd through Tm), three moment vectors per site were
considered: the conduction electron (spd) spin moment vector and the 4f electron spin and
orbital moment vectors. The eective Hamiltonian in the latter case contains also intrasite spin-
orbit coupling interaction and crystal-eld terms. A rst-principles approach to crystal-eld
parameters in RE-based systems can be found in Refs. [60, 108].
As mentioned in Section 2, the pair exchange interactions according to Eq. (5) do not contain
contributions from constraining magnetic elds which appear as Lagrange multipliers in the
constrained DFT. Their eect has been recently addressed in Refs. [24, 26]; it leads to a mod-
ication of the pair exchange interactions and related quantities. The changes of the moment
directions u
R
and the constraining elds B

R
satisfy linear relations [24]
M
R
u
R
=

R

_
K
RR
u
R
+
RR
B

_
, (36)
where M
R
denotes the magnitude of R-th local moment in units of
B
and where the exchange-
141
correlation response function K
RR
and the bare transverse susceptibility
RR
are given by
K
RR
=
2

Im
_
E
F

dE
_

R
dr
_

dr

(r, r

; E + i0) B
xc
(r

) G

(r

, r; E + i0) ,

RR
=
2

Im
_
E
F

dE
_

R
dr
_

dr

(r, r

; E + i0) G

(r

, r; E + i0) . (37)
As a consequence of non-zero constraining elds, the exchange parameters J
RR
get renormalized
values given explicitly by [24]
J
ren
RR
= J
RR

1
2
_
(M K
T
) X
1
(M K)
_
RR

, (38)
where M, K, X denote, respectively, matrices with elements M
R

RR
, K
RR
,
RR
, and where
K
T
denotes the transpose of K. The evaluation of Eq. (38) for real systems remains yet to be
performed.
Some physical insight into the nature of this renormalization can be obtained for Bravais lattices
assuming a suciently rigid magnetization within an atomic cell [24]. With a denition of an
average exchange splitting on site R,

R
=
2
M
R
_

R
dr m(r) B
xc
(r) , (39)
where m(r) is the spin density, it can be shown that the renormalization of the pair interactions,
Eq. (38), leads to a simple renormalization of the magnon energies E(q) which is signicant
for high-energy magnons E(q) > (where =
R
for a Bravais lattice). In particular, the
spin-wave stiness constant D does not undergo any renormalization. The Curie temperature
evaluated in the RPA becomes renormalized according to a formula
_
k
B
T
RPA,ren
C
_
1
=
_
k
B
T
RPA
C
_
1

6
M
, (40)
which means that the renormalized Curie temperatures are enhanced as compared to the un-
renormalized ones. The values for the bulk cubic 3d ferromagnets are summarized in Table 3. It
can be seen that the renormalization of exchange parameters improves considerably the agree-
ment between theory and experiment for Fe and Ni, in contrast to Co, where the unrenormalized
Curie temperature is closer to experiment than the renormalized one. A quantitative analysis
for bulk Gd shows that the eect of constraining elds is completely negligible [26].
Table 3: Curie temperature calculated within the RPA by using the bare (T
RPA
C
) and renormal-
ized (T
RPA,ren
C
) exchange interactions and their comparison with experimental values (T
C,exp
).
Metal T
RPA
C
[K] T
RPA,ren
C
[K] T
C,exp
[K]
Fe bcc 950 1057 1044 1045
Co fcc 1311 1771 1388 1398
Ni fcc 350 634 624 631
As is well known, the exact magnon dispersion law is obtained from poles of the dynamical
transverse susceptibility while the eective exchange interactions are closely related to the static
142
transverse susceptibility [26]. Susceptibility calculations within the DFT are, however, quite
involved even for systems with perfect three-dimensional translational symmetry [109, 110],
which calls for additional approximations. The calculations can be simplied using again the
adiabatic approximation and the ASA as done in Ref. [102]. The spin-wave spectrum of bcc Fe
calculated from the susceptibility agrees very well with that from a frozen-magnon approach,
whereas dierences between the two approaches are found in results for fcc Co and fcc Ni [102].
However, for a correct reproduction of the optical magnon branch of fcc Ni (including the
lifetime eects), full susceptibility calculations [110] are inevitable.
Susceptibility calculations have been employed in another successful approach to nite-tempe-
rature magnetism, based on the DLM state with local magnetic moments pointing to random
directions [21, 111]. In contrast to the previous techniques starting from the magnetic ground
state and its excitations, the DLM theory is focused on the paramagnetic state whereby no
particular form of an eective spin Hamiltonian has to be assumed. The transition temperature
is derived from divergence of the susceptibility of the DLM state which is related to a direct
correlation function. An ecient evaluation of the latter quantity represents the most dicult
part of computations. This mean-eld technique has been applied to bulk bcc Fe and fcc Ni [54];
an especially good agreement between the calculated and experimental Curie temperatures has
been achieved by including eects of correlations in terms of Onsager cavity elds. Applications
to thin transition-metal lms have been restricted to the mean-eld level [112, 113]; the results
for the Curie temperatures reproduce a number of experimentally observed trends with respect
to the thickness of magnetic lms and of non-magnetic capping layers. In the case of Fe and
Co monolayers on a Cu(001) substrate, however, the theory predicts unrealistically high Curie
temperatures, in full agreement with the MFA values from the EHH, see Section 4.5 and Ref. [13].
The most important feature of the described ab initio approach to exchange interactions lies in
its real-space formulation, which opens a way to study long-range interactions encountered in
itinerant magnets including systems without three-dimensional translational invariance (random
alloys, low-dimensional magnets). The inherent limitation to cases with large local magnetic
moments makes the approach especially suitable for applications to systems like transition-metal
surfaces and thin lms, diluted magnetic semiconductors, rare-earth metals and compounds, etc.
Acknowledgements
We are grateful to L. Bergqvist, G. Bihlmayer, S. Bl ugel, G. Bouzerar, M. Divis, P. Franek, F.
M aca, M. Pajda, J. Rusz and P. Weinberger for collaboration at various stages of the project,
and to P. Dederichs, B. Gyory, P. Levy, A. Liechtenstein, A. Shick, L. Szunyogh and L. Udvardi
for helpful discussions.
We acknowledge the nancial support provided by the Grant Agency of the Czech Republic (No.
106/02/0943), the Academy of Sciences of the Czech Republic (No. A1010203 and Z2041904), the
Ministry of Education of the Czech Republic (MSM113200002), the Center for Computational
Materials Science in Vienna (GZ 45.504), and the RT Network Computational Magnetoelec-
tronics (Contract HPRN-CT-2000-00143) of the European Commission.
143
References
[1] P. Hohenberg and W. Kohn. Phys. Rev., 136:B864, 1964.
[2] W. Kohn and L. J. Sham. Phys. Rev., 140:A1133, 1965.
[3] U. von Barth and L. Hedin. J. Phys. C, 5:1629, 1972.
[4] J. B. Staunton. Rep. Progr. Phys., 57:1289, 1994.
[5] J. K ubler. Theory of Itinerant Electron Magnetism. Clarendon Press, Oxford, 2000.
[6] K. Baberschke, M. Donath, and W. Nolting, editors. Band-Ferromagnetism, volume 580 of Lecture
Notes in Physics. Springer, Berlin, 2001.
[7] P. Mohn. Magnetism in the Solid State. Springer, Berlin, 2003.
[8] E. Runge and E. K. U. Gross. Phys. Rev. Lett., 52:997, 1984.
[9] N. D. Mermin. Phys. Rev., 137:A1441, 1965.
[10] A. I. Liechtenstein, M. I. Katsnelson, and V. A. Gubanov. J. Phys. F, 14:L125, 1984.
[11] A. I. Liechtenstein, M. I. Katsnelson, V. P. Antropov, and V. A. Gubanov. J. Magn. Magn. Mater.,
67:65, 1987.
[12] V. P. Antropov, B. N. Harmon, and A. N. Smirnov. J. Magn. Magn. Mater., 200:148, 1999.
[13] M. Pajda, J. Kudrnovsk y, I. Turek, V. Drchal, and P. Bruno. Phys. Rev. Lett., 85:5424, 2000.
[14] M. Pajda, J. Kudrnovsk y, I. Turek, V. Drchal, and P. Bruno. Phys. Rev. B, 64:174402, 2001.
[15] P. Bruno, J. Kudrnovsk y, M. Pajda, V. Drchal, and I. Turek. J. Magn. Magn. Mater., 240:346,
2002.
[16] J. Kudrnovsk y, I. Turek, V. Drchal, F. M aca, J. Masek, P. Weinberger, and P. Bruno. J. Super-
conductivity, 16:119, 2003.
[17] I. Turek, J. Kudrnovsk y, G. Bihlmayer, and S. Bl ugel. J. Phys.: Condens. Matter, 15:2771, 2003.
[18] I. Turek, J. Kudrnovsk y, V. Drchal, P. Bruno, and S. Bl ugel. Physica Status Solidi B, 236:318,
2003.
[19] I. Turek, G. Bihlmayer, S. Bl ugel, and P. Weinberger. Czech. J. Phys., 53:81, 2003.
[20] G. Bouzerar, J. Kudrnovsk y, L. Bergqvist, and P. Bruno. Phys. Rev. B, 68:081203, 2003.
[21] B. L. Gyory, A. J. Pindor, J. Staunton, G. M. Stocks, and H. Winter. J. Phys. F, 15:1337, 1985.
[22] P. H. Dederichs, S. Bl ugel, R. Zeller, and H. Akai. Phys. Rev. Lett., 53:2512, 1984.
[23] L. M. Sandratskii. J. Phys.: Condens. Matter, 3:8565, 1991.
[24] P. Bruno. Phys. Rev. Lett., 90:087205, 2003.
[25] A. Oswald, R. Zeller, P. J. Braspenning, and P. H. Dederichs. J. Phys. F, 15:193, 1985.
[26] V. P. Antropov. J. Magn. Magn. Mater., 262:L192, 2003.
[27] S. V. Halilov, H. Eschrig, A. Y. Perlov, and P. M. Oppeneer. Phys. Rev. B, 58:293, 1998.
[28] V. P. Antropov, M. I. Katsnelson, M. van Schilfgaarde, and B. N. Harmon. Phys. Rev. Lett., 75:729,
1995.
[29] V. P. Antropov, M. I. Katsnelson, B. N. Harmon, M. van Schilfgaarde, and D. Kusnezov. Phys.
Rev. B, 54:1019, 1996.
144
[30] Q. Niu and L. Kleinman. Phys. Rev. Lett., 80:2205, 1998.
[31] Q. Niu, X. Wang, L. Kleinman, W.-M. Liu, D. M. C. Nicholson, and G. M. Stocks. Phys. Rev.
Lett., 83:207, 1999.
[32] S. V. Tyablikov. Methods of Quantum Theory of Magnetism. Plenum Press, New York, 1967.
[33] C. S. Wang, R. E. Prange, and V. Korenman. Phys. Rev. B, 25:5766, 1982.
[34] O. K. Andersen and O. Jepsen. Phys. Rev. Lett., 53:2571, 1984.
[35] I. Turek, V. Drchal, J. Kudrnovsk y, M.

Sob, and P. Weinberger. Electronic Structure of Disordered
Alloys, Surfaces and Interfaces. Kluwer, Boston, 1997.
[36] I. Turek, J. Kudrnovsk y, and V. Drchal. In H. Dreysse, editor, Electronic Structure and Physical
Properties of Solids, volume 535 of Lecture Notes in Physics, page 349. Springer, Berlin, 2000.
[37] S. H. Vosko, L. Wilk, and M. Nusair. Can. J. Phys., 58:1200, 1980.
[38] P. Weinberger. Electron Scattering Theory for Ordered and Disordered Matter. Clarendon Press,
Oxford, 1990.
[39] A. Gonis. Green Functions for Ordered and Disordered Systems. North-Holland, Amsterdam, 1992.
[40] O. Gunnarsson, O. Jepsen, and O. K. Andersen. Phys. Rev. B, 27:7144, 1983.
[41] P. Bruno. Phys. Rev. B, 52:411, 1995.
[42] K. C. Hass, B. Velick y, and H. Ehrenreich. Phys. Rev. B, 29:3697, 1984.
[43] S. Halilov. In D. J. Singh and D. A. Papaconstantopoulos, editors, Electronic Structure and Mag-
netism of Complex Materials, volume 54 of Springer Series in Materials Science, chapter 1, page 1.
Springer, Berlin, 2003.
[44] M. van Schilfgaarde and V. P. Antropov. J. Appl. Phys., 85:4827, 1999.
[45] C. K. Loong, J. M. Carpenter, J. W. Lynn, R. A. Robinson, and H. A. Mook. J. Appl. Phys.,
55:1895, 1984.
[46] J. W. Lynn. Phys. Rev. B, 11:2624, 1975.
[47] H. A. Mook and D. McK. Paul. Phys. Rev. Lett., 54:227, 1985.
[48] G. Shirane, V. J. Minkiewicz, and R. Nathans. J. Appl. Phys., 39:383, 1968.
[49] H. A. Mook, J. W. Lynn, and M. R. Nicklow. Phys. Rev. Lett., 30:556, 1973.
[50] R. Pauthenet. J. Appl. Phys., 53:2029 and 8187, 1982.
[51] D. Spis ak and J. Hafner. J. Magn. Magn. Mater., 168:257, 1997.
[52] V. P. Antropov, M. I. Katsnelson, and A. I. Liechtenstein. Physica B, 237-238:336, 1997.
[53] R. H. Brown, D. M. C. Nicholson, X. Wang, and T. C. Schulthess. J. Appl. Phys., 85:4830, 1999.
[54] J. B. Staunton and B. L. Gyory. Phys. Rev. Lett., 69:371, 1992.
[55] M. Heinemann and W. M. Temmerman. Phys. Rev. B, 49:4348, 1994.
[56] J. Jensen and A. R. Mackintosh. Rare Earth Magnetism. Clarendon, Oxford, 1991.
[57] B. N. Harmon, V. P. Antropov, A. I. Liechtenstein, I. V. Solovyev, and V. I. Anisimov. J. Phys.
Chem. Solids, 56:1521, 1995.
[58] A. B. Shick, A. I. Liechtenstein, and W. E. Pickett. Phys. Rev. B, 60:10763, 1999.
[59] P. Strange, A. Svane, W. M. Temmerman, Z. Szotek, and H. Winter. Nature, 399:756, 1999.
145
[60] M. Richter. J. Phys. D, 31:1017, 1998.
[61] M. S. S. Brooks, L. Nordstr om, and B. Johansson. J. Phys.: Condens. Matter, 3:2357, 1991.
[62] A. Y. Perlov, S. V. Halilov, and H. Eschrig. Phys. Rev. B, 61:4070, 2000.
[63] P. Kurz, G. Bihlmayer, and S. Bl ugel. J. Phys.: Condens. Matter, 14:6353, 2002.
[64] P. A. Lindgard. Phys. Rev. B, 17:2348, 1978.
[65] N. G. Nereson, C. E. Olsen, and G. P. Arnold. Phys. Rev., 135:A176, 1964.
[66] A. H. Millhouse and K. A. McEwen. Solid State Commun., 13:339, 1973.
[67] P. Franek. Masters thesis, Charles University, Prague, 2003.
[68] M. S. S. Brooks and B. Johansson. In K. H. J. Buschow, editor, Handbook of Magnetic Materials,
volume 7, chapter 3, page 139. North-Holland, Amsterdam, 1993.
[69] I. Turek, J. Kudrnovsk y, M. Divis, P. Franek, G. Bihlmayer, and S. Bl ugel. Phys. Rev. B, (submit-
ted).
[70] J. Kudrnovsk y and V. Drchal. Phys. Rev. B, 41:7515, 1990.
[71] P. Bruno, J. Kudrnovsk y, V. Drchal, and I. Turek. Phys. Rev. Lett., 76:4254, 1996.
[72] J. Kudrnovsk y, V. Drchal, I. Turek, P. Bruno, P. Dederichs, and P. Weinberger. In H. Dreysse,
editor, Electronic Structure and Physical Properties of Solids, volume 535 of Lecture Notes in
Physics, page 313. Springer, Berlin, 2000.
[73] P. M. Levy, S. Maekawa, and P. Bruno. Phys. Rev. B, 58:5588, 1998.
[74] S. D. Drell and L. Verlet. Phys. Rev., 99:849, 1955.
[75] G. Bouzerar and P. Bruno. Phys. Rev. B, 66:014410, 2002.
[76] S. Ghosh, P. L. Leath, and M. H. Cohen. Phys. Rev. B, 66:214206, 2002.
[77] H. Akai. Phys. Rev. Lett., 81:3002, 1998.
[78] H. Ohno. J. Magn. Magn. Mater., 200:110, 1999.
[79] K. W. Edmonds et al. Appl. Phys. Lett., 81:4991, 2002.
[80] F. M aca and J. Masek. Phys. Rev. B, 65:235209, 2002.
[81] P. A. Korzhavyi et al. Phys. Rev. Lett., 88:187202, 2002.
[82] K. Sato, P. H. Dederichs, and H. Katayama-Yoshida. Europhys. Lett., 61:403, 2003.
[83] L. M. Sandratskii and P. Bruno. Phys. Rev. B, 66:134435, 2002.
[84] Y. D. Park et al. Science, 295:651, 2002.
[85] B. T. Jonker et al. cond-mat/0302231.
[86] M. Weinert and S. Bl ugel. In L. H. Bennett and R. E. Watson, editors, Magnetic Multilayers,
page 51. World Scientic, Singapore, 1994.
[87] R. Vollmer, S. van Dijken, M. Schleberger, and J. Kirschner. Phys. Rev. B, 61:1303, 2000.
[88] N. D. Mermin and H. Wagner. Phys. Rev. Lett., 17:1133, 1966.
[89] P. Bruno. Mater. Res. Soc. Symp. Proc., 321:299, 1992.
[90] A. Ney, F. Wilhelm, M. Farle, P. Poulopoulos, P. Srivastava, and K. Baberschke. Phys. Rev. B,
59:R3938, 1999.
146
[91] E. I. Isaev, L. V. Pourovskii, A. M. N. Niklasson, Yu. Kh. Vekilov, B. Johansson, and I. A.
Abrikosov. Phys. Rev. B, 65:024435, 2001.
[92] E. D. Tober et al. Phys. Rev. Lett., 81:2360, 1998.
[93] A. B. Shick, W. E. Pickett, and C. S. Fadley. Phys. Rev. B, 61:R9213, 2000.
[94] C. S. Arnold and D. P. Pappas. Phys. Rev. Lett., 85:5202, 2000.
[95] S. Bl ugel. In Magnetismus von Festk orpern und Grenz achen. Forschungszentrum J ulich, 1993.
[96] J. Giergiel, A. W. Pang, H. Hopster, X. Guo, S. Y. Tong, and D. Weller. Phys. Rev. B, 51:10201,
1995.
[97] D. Spis ak and J. Hafner. Phys. Rev. B, 65:235405, 2002.
[98] D. K odderitzsch, W. Hergert, W. M. Temmerman, Z. Szotek, A. Ernst, and H. Winter. Phys. Rev.
B, 66:064434, 2002.
[99] R. F. Sabiryanov, S. K. Bose, and O. N. Mryasov. Phys. Rev. B, 51:8958, 1995.
[100] S. Frota-Pessoa, R. B. Muniz, and J. Kudrnovsk y. Phys. Rev. B, 62:5293, 2000.
[101] L. Szunyogh and L. Udvardi. J. Magn. Magn. Mater., 198-199:537, 1999.
[102] O. Grotheer, C. Ederer, and M. F ahnle. Phys. Rev. B, 63:100401(R), 2001.
[103] O. N. Mryasov, A. J. Freeman, and A. I. Liechtenstein. J. Appl. Phys., 79:4805, 1996.
[104] V. P. Antropov and A. I. Liechtenstein. Mater. Res. Soc. Symp. Proc., 253:325, 1992.
[105] L. Udvardi, L. Szunyogh, K. Palot as, and P. Weinberger. Phys. Rev. B, 68:1044XX, 2003.
[106] M. I. Katsnelson and A. I. Lichtenstein. Phys. Rev. B, 61:8906, 2000.
[107] N. M. Rosengaard and B. Johansson. Phys. Rev. B, 55:14975, 1997.
[108] P. Nov ak and J. Kuriplach. Phys. Rev. B, 50:2085, 1994.
[109] J. Callaway, A. K. Chatterjee, S. P. Singhal, and A. Ziegler. Phys. Rev. B, 28:3818, 1983.
[110] S. Y. Savrasov. Phys. Rev. Lett., 81:2570, 1998.
[111] J. B. Staunton, S. S. A. Razee, L. Szunyogh, and B. L. Gyory. Physica B, 318:316, 2002.
[112] S. S. A. Razee, J. B. Staunton, L. Szunyogh, and B. L. Gyory. Phys. Rev. B, 66:094415, 2002.
[113] S. S. A. Razee, J. B. Staunton, L. Szunyogh, and B. L. Gyory. Phys. Rev. Lett., 88:147201, 2002.
147